In [1]:
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import warnings
from tabulate import tabulate
import plotly.express as px
import plotly.graph_objects as go
from sklearn.datasets import load_iris
from mlxtend.plotting import plot_decision_regions
from sklearn.model_selection import train_test_split
from sklearn.model_selection import cross_val_score
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score
from sklearn.preprocessing import LabelEncoder
Loading the Dataset¶
In [2]:
# Load the Iris dataset
iris = load_iris(as_frame=True)
# Convert data and target attributes to DataFrame
iris_df = pd.concat([iris.data, iris.target], axis=1)
iris_df.columns = iris.feature_names + ['target']
# Replace numerical target values with class names
iris_df['target'] = iris.target_names[iris_df['target']]
# Display the DataFrame
print(iris_df)
sepal length (cm) sepal width (cm) petal length (cm) petal width (cm)
0 5.1 3.5 1.4 0.2 \
1 4.9 3.0 1.4 0.2
2 4.7 3.2 1.3 0.2
3 4.6 3.1 1.5 0.2
4 5.0 3.6 1.4 0.2
.. ... ... ... ...
145 6.7 3.0 5.2 2.3
146 6.3 2.5 5.0 1.9
147 6.5 3.0 5.2 2.0
148 6.2 3.4 5.4 2.3
149 5.9 3.0 5.1 1.8
target
0 setosa
1 setosa
2 setosa
3 setosa
4 setosa
.. ...
145 virginica
146 virginica
147 virginica
148 virginica
149 virginica
[150 rows x 5 columns]
In [3]:
iris_df["target"]
Out[3]:
0 setosa
1 setosa
2 setosa
3 setosa
4 setosa
...
145 virginica
146 virginica
147 virginica
148 virginica
149 virginica
Name: target, Length: 150, dtype: object
Redefine Classes Into: virginica | non-virginica¶
In [4]:
for i in range(0, len(iris_df["target"])):
if iris_df["target"][i] == 'virginica':
iris_df.loc[i, "target"] = 'virginica'
else:
iris_df.loc[i, "target"] = 'non-virginica'
Descriptive Statistics¶
Virginica Group¶
In [5]:
virginica_group = iris_df[iris_df.target == 'virginica']
virginica_group.head()
Out[5]:
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | target | |
|---|---|---|---|---|---|
| 100 | 6.3 | 3.3 | 6.0 | 2.5 | virginica |
| 101 | 5.8 | 2.7 | 5.1 | 1.9 | virginica |
| 102 | 7.1 | 3.0 | 5.9 | 2.1 | virginica |
| 103 | 6.3 | 2.9 | 5.6 | 1.8 | virginica |
| 104 | 6.5 | 3.0 | 5.8 | 2.2 | virginica |
In [6]:
virginica_group.describe()
Out[6]:
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | |
|---|---|---|---|---|
| count | 50.00000 | 50.000000 | 50.000000 | 50.00000 |
| mean | 6.58800 | 2.974000 | 5.552000 | 2.02600 |
| std | 0.63588 | 0.322497 | 0.551895 | 0.27465 |
| min | 4.90000 | 2.200000 | 4.500000 | 1.40000 |
| 25% | 6.22500 | 2.800000 | 5.100000 | 1.80000 |
| 50% | 6.50000 | 3.000000 | 5.550000 | 2.00000 |
| 75% | 6.90000 | 3.175000 | 5.875000 | 2.30000 |
| max | 7.90000 | 3.800000 | 6.900000 | 2.50000 |
Non-Virginica Group¶
In [7]:
non_virginica_group = iris_df[iris_df.target == 'non-virginica']
non_virginica_group.head()
Out[7]:
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | target | |
|---|---|---|---|---|---|
| 0 | 5.1 | 3.5 | 1.4 | 0.2 | non-virginica |
| 1 | 4.9 | 3.0 | 1.4 | 0.2 | non-virginica |
| 2 | 4.7 | 3.2 | 1.3 | 0.2 | non-virginica |
| 3 | 4.6 | 3.1 | 1.5 | 0.2 | non-virginica |
| 4 | 5.0 | 3.6 | 1.4 | 0.2 | non-virginica |
In [8]:
non_virginica_group.describe()
Out[8]:
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | |
|---|---|---|---|---|
| count | 100.000000 | 100.000000 | 100.000000 | 100.000000 |
| mean | 5.471000 | 3.099000 | 2.861000 | 0.786000 |
| std | 0.641698 | 0.478739 | 1.449549 | 0.565153 |
| min | 4.300000 | 2.000000 | 1.000000 | 0.100000 |
| 25% | 5.000000 | 2.800000 | 1.500000 | 0.200000 |
| 50% | 5.400000 | 3.050000 | 2.450000 | 0.800000 |
| 75% | 5.900000 | 3.400000 | 4.325000 | 1.300000 |
| max | 7.000000 | 4.400000 | 5.100000 | 1.800000 |
Histogram for Each Feature.¶
In [9]:
# Plot histograms for each feature, separated by class
for feature in iris_df.columns[:-1]: # Exclude the target column
plt.figure(figsize=(8, 6))
sns.histplot(data=iris_df, x=feature, hue="target", hue_order=['non-virginica', 'virginica'], kde=True, legend=True)
plt.title(f"Histogram of {feature} for each class")
plt.xlabel(feature)
plt.ylabel("Frequency")
plt.show()
From the various histograms for each feature across the virginica and non-virginica group, we can analyze the relations and this helps get more insights from the dataset and clearly demonstrate the key differences between the two groups.
Non-Virginica group:
- Most have very low petal width and length among the iris flowers
Virginica group:
- They usually have higher petal width and length among the iris flowers
Sepal Length and width is observed to be evenly distributed across both iris groups.
Correlation Matrix¶
Virginica, Non-Virginica Separately¶
In [10]:
# Computing the correlation matrices
corr_matrix_virginica = virginica_group.iloc[:, :-1].corr()
corr_matrix_non_virginica = non_virginica_group.iloc[:, :-1].corr()
fig, axes = plt.subplots(1, 2, figsize=(16, 6))
# Virginica
sns.heatmap(corr_matrix_virginica, annot=True, cmap='coolwarm', ax=axes[0])
axes[0].set_title('Correlation Matrix of Features for the "virginica" group')
# Non-Virginica
sns.heatmap(corr_matrix_non_virginica, annot=True, cmap='coolwarm', ax=axes[1])
axes[1].set_title('Correlation Matrix of Features for the "non-virginica" group')
plt.tight_layout()
plt.show()
We can see that there is a high correlation between the petal lenght and sepal length in the Virginica Group.
There is a significantly high correlation between the petal lenght and petal width for the non-virginica group, which can be considered as a key factor to differenciate between Virginica and Non-Virginica groups.
Combined Dataset¶
In [11]:
# Computing the correlation matrix
corr_matrix = iris_df.iloc[:, :-1].corr()
sns.heatmap(corr_matrix, annot=True, cmap='coolwarm')
plt.title('Correlation Matrix of Features')
plt.show()
- This gives an overall view about the relationships between the attributes when conisdering the entire Iris Dataset without any classifications.
- We can notice that the petal length - petal width, petal length - sepal lenght have relations.
- This information can be considered when trying to differentiate Iris flowers from other flowers in general.
In [12]:
fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(6, 6))
iris_features = iris_df.columns[:-1].to_list()
for i, ax in enumerate(axes.flatten()):
sns.violinplot(data=iris_df, x="target", y=iris_df[iris_features[i]], palette="viridis", ax=ax)
plt.figtext(0, 0, "Reference: Kaggle - MURILÃO (Basic visualization techniques)")
plt.tight_layout()
plt.show()
Box Plots¶
In [13]:
fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(14, 10))
# Loop through and creating boxplots
for i, ax in enumerate(axes.flatten()):
sns.boxplot(x='target', y=iris_df.columns[i], data=iris_df, ax=ax)
plt.figtext(0, 0, "Reference: Kaggle - SAURAV ANAND (Seaborn Tutorial)")
plt.tight_layout()
plt.show()
Referenced: https://www.kaggle.com/code/saurav9786/seaborn-tutorial
We can see that when the sepal lengths are considered: Virginica Group has an outlier while when consider sepal width, both Virginica and non-Virginica group have single outlier each. This can impact the model's prediction by a bit which can be checked while training and evaluation of models.
Andrew Curves¶
In [14]:
# Create Andrews curves
pd.plotting.andrews_curves(iris_df, 'target', colormap='viridis')
plt.figtext(0, 0, "Reference: Kaggle - BEN HAMNER (Python Data Visualizations)")
plt.title("Andrews Curves of Iris Dataset")
plt.legend(title='Species', loc='upper right')
plt.show()
Referenced: https://www.kaggle.com/code/benhamner/python-data-visualizations
These offer a concise visualization of multivariate data as they enables easy identification of underlying structures without extensive statistical analysis.
The following can be analyzed:
- Both species' curves demonstrate consistent wave-like patterns with similar peak levels and few similar features.
- There is a clear differences between both curves. Non-virginica curves generally exhibit lower values compared to virginica curves.
Parallel Curves¶
In [15]:
pd.plotting.parallel_coordinates(iris_df, "target")
plt.figtext(0, 0, "Reference: Kaggle - BEN HAMNER (Python Data Visualizations)")
plt.title("Parallel Coordinates of Iris Dataset")
plt.legend(title='Species', loc='upper right')
plt.show()
Referenced: https://www.kaggle.com/code/benhamner/python-data-visualizations
Parallel Coordinates:¶
Provide clear relationships within multivariate data by allowing us to do a straightforward comparison of multiple variables across different species, identifying distinct patterns and trends.
We can analyze that:
Non-virginica generally exhibits wider sepal widths but shorter petals compared to virginica.
Virginica species tend to have longer and narrower sepals alongside longer petals.
Splitting the Data into Train, Validation and Test¶
In [16]:
# Splitting the data into 120 records for train and 30 records for test initially
X_train, X_temp, y_train, y_temp = train_test_split(iris_df.iloc[:, :-1], iris_df['target'], test_size=0.2, random_state=42)
# Split test data into 15 records for validation, 15 records for test
X_val, X_test, y_val, y_test = train_test_split(X_temp, y_temp, test_size=0.5, random_state=42)
# Display the shapes of the resulting sets
print("Train set shape:", X_train.shape, y_train.shape)
print("Validation set shape:", X_val.shape, y_val.shape)
print("Test set shape:", X_test.shape, y_test.shape)
Train set shape: (120, 4) (120,) Validation set shape: (15, 4) (15,) Test set shape: (15, 4) (15,)
In [17]:
warnings.filterwarnings("ignore")
def train_and_evaluate(X_train, X_val, X_test, y_train, y_val, y_test, num_features):
# Select the columns for training
X_train_subset = X_train.iloc[:, :num_features]
X_val_subset = X_val.iloc[:, :num_features]
X_test_subset = X_test.iloc[:, :num_features]
# Initialize and train logistic regression model
model = LogisticRegression()
model.fit(X_train_subset, y_train)
# Make predictions on validation set
y_pred_val = model.predict(X_val_subset)
# Calculate accuracy on validation set
accuracy_val = accuracy_score(y_val, y_pred_val)
# Make predictions on test set
y_pred_test = model.predict(X_test_subset)
# Calculate accuracy on test set
accuracy_test = accuracy_score(y_test, y_pred_test)
return accuracy_val, accuracy_test
# Train and evaluate logistic regression models
feature_names = iris.feature_names
for num_features in range(1, 5):
accuracy_val, accuracy_test = train_and_evaluate(X_train, X_val, X_test, y_train, y_val, y_test, num_features)
print(f"Model with {num_features} feature:")
print(f"Features: {', '.join(feature_names[:num_features])}")
print(f"Validation accuracy: {accuracy_val:.2f}")
print(f"Test accuracy: {accuracy_test:.2f}\n")
Model with 1 feature: Features: sepal length (cm) Validation accuracy: 0.93 Test accuracy: 0.93 Model with 2 feature: Features: sepal length (cm), sepal width (cm) Validation accuracy: 0.93 Test accuracy: 0.87 Model with 3 feature: Features: sepal length (cm), sepal width (cm), petal length (cm) Validation accuracy: 1.00 Test accuracy: 1.00 Model with 4 feature: Features: sepal length (cm), sepal width (cm), petal length (cm), petal width (cm) Validation accuracy: 1.00 Test accuracy: 1.00
In [18]:
def cross_validate(X, y, num_features):
# Selecting the columns for training
X_subset = X.iloc[:, :num_features]
# Initialize logistic regression model
model = LogisticRegression()
# Perform 5-fold cross-validation
scores = cross_val_score(model, X_subset, y, cv=5)
return scores.mean()
# Perform cross-validation for each number of features
for num_features in range(1, 5):
mean_accuracy = cross_validate(pd.concat([X_train, X_val]), pd.concat([y_train, y_val]), num_features)
print(f"Model with {num_features} feature(s) - Cross-validation accuracy: {mean_accuracy:.4f}")
Model with 1 feature(s) - Cross-validation accuracy: 0.8074 Model with 2 feature(s) - Cross-validation accuracy: 0.7926 Model with 3 feature(s) - Cross-validation accuracy: 0.9556 Model with 4 feature(s) - Cross-validation accuracy: 0.9630
- By performing cross validation across the various models, we notice that the models' performance has decreased, which can be a indication towards the overfitting.
- We can also observe that the models with features 1 and 2 perform substantially less when compared to 3 and 4.
- The possibility for this can both be that there are a few outliers present in the data.
- Usually, overall performance is less when considering fewer features for the model.
Evaluate Model's Performance in Tables¶
In [19]:
def create_evaluation_table(model, X_val, y_val):
# Make predictions and probabilities
y_pred = model.predict(X_val)
y_pred_proba = model.predict_proba(X_val)
# Get the occurrence number from the original dataset
occurrence_number = X_val.index + 1
# Creating the evaluation table
evaluation_table = pd.DataFrame({
'Instance Number': occurrence_number,
'Probability of Virginica': y_pred_proba[:, 1],
'Prediction': y_pred,
'Ground Truth': y_val
})
return evaluation_table.set_index('Instance Number')
# Define a function to print the evaluation table for a given model
def print_evaluation_table(model_name, evaluation_table):
print(f"Evaluation Table for Model: {model_name}")
print(tabulate(evaluation_table, headers='keys', tablefmt='grid'))
print()
# Iterate through each model and print its evaluation table
def individual_table(num_features):
# Training the model
model = LogisticRegression()
model.fit(X_train.iloc[:, :num_features], y_train)
# Creating the evaluation table
evaluation_table = create_evaluation_table(model, X_val.iloc[:, :num_features], y_val)
print_evaluation_table(f"Model with {num_features} feature(s)", evaluation_table)
return evaluation_table
def calculate_accuracy(evaluation_table):
# Count number of correct predictions
correct_predictions = (evaluation_table['Prediction'] == evaluation_table['Ground Truth']).sum()
# Total instances
total_instances = len(evaluation_table)
# Calculate accuracy
accuracy = (correct_predictions / total_instances) * 100
print(f"{accuracy:.2f}% is the prediction accuracy.")
In [20]:
predict_table = individual_table(1)
calculate_accuracy(predict_table)
Evaluation Table for Model: Model with 1 feature(s) +-------------------+----------------------------+---------------+----------------+ | Instance Number | Probability of Virginica | Prediction | Ground Truth | +===================+============================+===============+================+ | 27 | 0.06451 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 19 | 0.217912 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 119 | 0.937717 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 146 | 0.671933 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 79 | 0.336388 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 128 | 0.382264 | non-virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 109 | 0.671933 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 56 | 0.217912 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 31 | 0.0442258 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 30 | 0.0365199 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 142 | 0.753228 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 111 | 0.578831 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 20 | 0.0776461 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 133 | 0.529589 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 65 | 0.185827 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ 93.33% is the prediction accuracy.
In [21]:
predict_table = individual_table(2)
calculate_accuracy(predict_table)
Evaluation Table for Model: Model with 2 feature(s) +-------------------+----------------------------+---------------+----------------+ | Instance Number | Probability of Virginica | Prediction | Ground Truth | +===================+============================+===============+================+ | 27 | 0.050796 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 19 | 0.145047 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 119 | 0.949865 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 146 | 0.66928 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 79 | 0.347378 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 128 | 0.379491 | non-virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 109 | 0.732581 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 56 | 0.237155 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 31 | 0.0412891 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 30 | 0.0321397 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 142 | 0.739476 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 111 | 0.546099 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 20 | 0.0487707 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 133 | 0.556691 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 65 | 0.193357 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ 93.33% is the prediction accuracy.
In [22]:
predict_table = individual_table(3)
calculate_accuracy(predict_table)
Evaluation Table for Model: Model with 3 feature(s) +-------------------+----------------------------+---------------+----------------+ | Instance Number | Probability of Virginica | Prediction | Ground Truth | +===================+============================+===============+================+ | 27 | 1.57225e-05 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 19 | 1.3357e-05 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 119 | 0.998416 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 146 | 0.697669 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 79 | 0.227787 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 128 | 0.520907 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 109 | 0.95854 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 56 | 0.262784 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 31 | 1.96817e-05 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 30 | 1.98174e-05 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 142 | 0.585897 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 111 | 0.622702 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 20 | 8.92127e-06 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 133 | 0.921539 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 65 | 0.0152908 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ 100.00% is the prediction accuracy.
In [23]:
predict_table = individual_table(4)
calculate_accuracy(predict_table)
Evaluation Table for Model: Model with 4 feature(s) +-------------------+----------------------------+---------------+----------------+ | Instance Number | Probability of Virginica | Prediction | Ground Truth | +===================+============================+===============+================+ | 27 | 8.96424e-06 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 19 | 5.70191e-06 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 119 | 0.998534 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 146 | 0.873922 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 79 | 0.207005 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 128 | 0.57273 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 109 | 0.946564 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 56 | 0.17067 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 31 | 7.70235e-06 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 30 | 7.52261e-06 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 142 | 0.820222 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 111 | 0.728198 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 20 | 4.12879e-06 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ | 133 | 0.956114 | virginica | virginica | +-------------------+----------------------------+---------------+----------------+ | 65 | 0.0161961 | non-virginica | non-virginica | +-------------------+----------------------------+---------------+----------------+ 100.00% is the prediction accuracy.
The prediction values in the data among all the tables can be summarized easily by using the accuracy which is calculated by taking the number of correct predictions and dividing it by total number of predictions made (Valid Prediction / Total Predictions) and then converting this into a percentage to make it more readable.
Since both the model with three features and the model with four features achieved 100% accuracy on the validation set, I would likely choose either of these models for making predictions on new data. Using fewer features may offer computational advantages and could result in a simpler and more interpretable model. Therefore, I would simply choose the model with three features for its computational efficiency and simplicity.
Re-Training¶
In [24]:
X_train, X_temp, y_train, y_temp = train_test_split(iris_df.iloc[:, :-1], iris_df['target'], test_size=0.2, random_state=42)
X_val, X_test, y_val, y_test = train_test_split(X_temp, y_temp, test_size=0.5, random_state=42)
Plotting the Decision Boundary¶
Modeles with Feature 1 and 2¶
In [25]:
iris_df.columns[1]
Out[25]:
'sepal width (cm)'
In [26]:
def train_plot_evaluate(X_train, X_val, y_train, y_val):
fig, axes = plt.subplots(1, 2, figsize=(12, 4))
for num_features, ax in zip(range(1, 3), axes):
# Training the model
model = LogisticRegression()
model.fit(X_train.iloc[:, :num_features], y_train)
# Convert X_val to NumPy array
X_val_np = X_val.iloc[:, :num_features].values
# Plot decision boundary
plot_decision_regions(X_val_np, y_val, clf=model, legend=2, ax=ax)
if num_features == 2:
ax.set_xlabel(iris_df.columns[num_features-1])
else:
ax.set_xlabel(iris_df.columns[num_features])
ax.set_ylabel(iris_df.columns[num_features])
ax.set_title(f"Decision Boundary with {num_features} Feature(s)")
# Modify legend
handles, labels = ax.get_legend_handles_labels()
ax.legend(handles, ["Non-Virginica", "Virginica"])
plt.tight_layout()
plt.show()
# Encode target labels
le = LabelEncoder()
y_train_encoded = le.fit_transform(y_train)
y_val_encoded = le.transform(y_val)
# Train, plot decision boundaries
train_plot_evaluate(X_train, X_val, y_train_encoded, y_val_encoded)
In [27]:
iris_df.columns
Out[27]:
Index(['sepal length (cm)', 'sepal width (cm)', 'petal length (cm)',
'petal width (cm)', 'target'],
dtype='object')
3D Plot for Model with 3 Features¶
In [28]:
def plot_decision_boundary_3d(X, y, model):
# Train the model
model.fit(X, y)
# Create a meshgrid for the features
feature1_vals = X[:, 0]
feature2_vals = X[:, 1]
feature3_vals = X[:, 2]
x_min, x_max = feature1_vals.min() - 1, feature1_vals.max() + 1
y_min, y_max = feature2_vals.min() - 1, feature2_vals.max() + 1
z_min, z_max = feature3_vals.min() - 1, feature3_vals.max() + 1
xx, yy, zz = np.meshgrid(np.arange(x_min, x_max, 0.1),
np.arange(y_min, y_max, 0.1),
np.arange(z_min, z_max, 0.1))
# Predict the labels for the meshgrid points
Z = model.predict(np.c_[xx.ravel(), yy.ravel(), zz.ravel()])
Z = Z.reshape(xx.shape)
# Plot decision boundary
fig = px.scatter_3d(x=feature1_vals, y=feature2_vals, z=feature3_vals, color=y)
fig.add_trace(go.Surface(x=xx.squeeze(), y=yy.squeeze(), z=zz.squeeze(),
surfacecolor=Z.squeeze(), colorscale='Viridis', opacity=0.5,
showscale=False))
# Plot decision boundary line
coef = model.coef_.squeeze()
intercept = model.intercept_
x_plane = np.linspace(x_min, x_max, 10)
y_plane = np.linspace(y_min, y_max, 10)
xx_plane, yy_plane = np.meshgrid(x_plane, y_plane)
z_plane = (-coef[0] * xx_plane - coef[1] * yy_plane - intercept) / coef[2]
fig.add_trace(go.Surface(x=xx_plane, y=yy_plane, z=z_plane,
opacity=0.5, showscale=False))
fig.update_layout(scene=dict(
xaxis_title='sepal length (cm)',
yaxis_title='sepal width (cm)',
zaxis_title='petal length (cm)'),
title='Decision Boundary in 3D')
fig.show()
# Encode target labels
le = LabelEncoder()
y_train_encoded = le.fit_transform(y_train)
# Convert DataFrame to numpy array
X_train_np = X_train.to_numpy()
model = LogisticRegression()
model.fit(X_train_np[:, :3], y_train_encoded)
plot_decision_boundary_3d(X_train_np[:, :3], y_train_encoded, model)